Some ideas

[marcharper]

• would be neat to apply classification algorithms (knn, svm, etc) on the fingerprint plots, just to separate the red and blue clusters. Are the linearly separable fingerprints similar?
• similarity scores e.g by % overlapping areas
• a series of fingerprints for a parameterized strategy like GTFT as the parameters increase
• self-dual strategies?
• Is there any relationship between fingerprints and Moran process outcomes? I.e. similar fingerprints are close to neutral fixations?
• higher-order fingerprints, e.g. for rock-paper-scissors

[drvinceknight]

Great ideas!

would be neat to apply classification algorithms (knn, svm, etc) on the fingerprint plots, just to separate the red and blue clusters. Are the linearly separable fingerprints similar?

Could actually do this on the underlying data (available from the fingerprint class).

similarity scores e.g by % overlapping areas a series of fingerprints for a parameterized strategy like GTFT as the parameters increase

This immediately made me realise that the fingerprints aren't implemented perfectly: they currently fingerprint a class which is a silly mistake. I've opened: https://github.com/Axelrod-Python/Axelrod/pull/785

self-dual strategies?

Strategies that are their own dual? Ashlock mentions this in "Fingerprinting: Visualization and Automatic Analysis of Prisoner’s Dilemma Strategies": Pavlov/WSLS is self dual. Would be interesting to look to see if there are others.

Is there any relationship between fingerprints and Moran process outcomes? I.e. similar fingerprints are close to neutral fixations?

Worth looking at.